Puja Ohlhaver, Mikhail Nikulin - Compressed to 0: The Silent Strings of Proof of Personhood

Transcript

great thank you for that introduction real full house for this talk um so generative AI raises a number of questions and anxieties will it replace our jobs can we trust what's real how do we govern it and some these are questions around Democratic governance economic distribution and the authenticity of information which is essential to both well functioning Market and also well functioning politics like democracy now this has spurred the rise of proof of personhood protocols which aims to verify unique humans Paving the way for Global Democratic processes Ubi

and even AI governance now the ambition of these protocols is simple represent each unique human with a corresponding unique digital identity one to one now today we want to offer a critique based off of the experience of idena protocol in their experiment and a paper we released just a few weeks ago entitled compressed to zero but first why idena well idena was the first proof of personhood blockchain with an endogenous network of validators with a one person onebot consensus algorithm moreover they succeeded in actually filtering Bots and validating humans and moreover they styi this problem of account trading with

unique novel mechanisms and account trading is where participants validate their accounts and then sell them off in a Black Market which is a problem which plagues many proof of personhood protocols today including worldcoin so idena had solved these problems now I will pass the mic to my esteemed co-author and founder of idena protocol Misha and he will explain the experiment to you thanks for joh so first of all why Idina was started um Idina aimed to address the challenge of proof of stake decentralization we all know that capital is always heavy tail par distributed and large staking pools have inevitable economy of scale which lead

that the fact that solar miners are Vanishing giving the way to Al legopolis also Adena aimed to solve a blockchain scalability thma like a blockchain scalability problem uh by establishing a network decentralized network of validators with excessive number of operators and fac facilitating sharding with parallel uh transactions and as synchronous execution with parallel State machines every note uh was intended to be operated by why one unique uh human with unique account so various proof of person protocols establish uniqueness in different ways unlike of w coin and

other protocols that relies on biometric data I relies on uh online simultaneous cognitive tests it's called so-called validation ceremonies which happen every few weeks synchronously for all participants so it's based on idea that you cannot be in multiple places at the same time and by solving the flip tests you actually prove that you are not bot and the flip test is a test based on the user generated story which makes sense in one configuration and it meaningless in another one so for example you can see the left sequence of images tell a cerent story uh about beasting whereas the casuality on the right side is

broken oops so what happened Aden uh launched in 2019 accounts steadily increased but as the network grew onchain data began to show some suspicious P patterns of transactions sending rewards from validated accounts to the same addresses at the same time and this kind of coordinated one-way transfers at the same time and the same wallets implied automation which requires third party key access to the participants accounts so either these participants were unwittingly seed their private key to third parties or they just never had them by December 2020 one of the users

contacted us and uh admitted uh to run uh the largest human Farm in the network so in fact High information operators pay low information participants like pets to periodically perform validation ceremony and verify their uniqueness in exchange for controlling their accounts and their private keys so puppets were either unaware about their private keys or they knew their private keys but were unaware about significance of the private Keys within the protocol so petering was not a traditional the jury Cil attack where malicious actors could fake multiple Cil accounts typically using some algorithmic Bots but it was instead as the protocol filterate BS and

authenticated U Flesh and Blood humans it was def facto like a uh Cil attack of humans acting as a programmable programmable bots so at best Peter Peters extracted some amount Delta between uh Ubi paid by the protocol and the wages paid to workers at worst they could just take all the rewards of the puppet accounts but was the picture so black and white uh pool operators also claimed that some awakened puppet were coming back so they were just offering a service to consensual participants and being

remunerated earning rewards on idena is presented with number of hustles so you need to run your mining note you need to um exchange Idina coin to the local currency and all these hustles could be operated by um could be kind of coordinated by the pool operators better than participants on their own in short they were cating and simply providing the service for some of the participants so in reality there are two extreme possibilities in the Spectrum in between like the same transaction pattern on chain with this one transfer to the same wallet to the same time could indicate like both petering or voluntary cation and both extremes different not in how

they looks like onchain but in the distribution of information and control offchain so in case of petering operators have more control and more information so they know how the protocol Works they know they control the private keys but uh and and by doing that they capitalize their as symmetry information and control in exchange for minimum wage payments and Maxim maximizing like their value and extracting value um from the protocol payouts in contrast ceration has greater symmetry information and control so participants know how the protocol Works they control their private keys but find it mutually beneficial to pool resources

and delegate control including their private uh private keys to operator and for greater rewards with economy of scale but at the same time keeping po operators accountable of chain so despite of uh establishing a transparent network with individuals with unique accounts the protocol collapsed into hidden groups and uh hidden sub networks these groups whether autocratic or Democratic were competing to control over fixed economic p in March 2021 uh 21 rather than just fighting pools and slushing the suspicious accounts ad Community uh agreed like voted for the fork and

agreed to bring the pools out out of the unquantified white shadows um the onchain delegation was introduced um uh this delegation uh gave pools economic incentives so ped account could earn mining rewards using only one note instead of multiple nodes for every account in the pool but what's more important with delegation pool operators could handle operational hustles uh and earn accounts rewards without needing to know the participants private keys so the delegation succeeded in uh making

pools transparent uh the green area shows uh the network pred delegation when accounts could not be uh differentiated uh uh when accounts uh even if accounts were controlled by the same pool operator and after delegation uh the protocol could measure the size of the pools so as the network grew the growth was notably among large pools which are red and pink while solar accounts which are blue at lined constituting smaller and smaller proportion of the network up until the peak of the peterian crises in May 2022 the percentage of the solar counts in blue shrunk from 62% to 27% of the network large pools had revers Trend ballooning from 22% to 61% of the

network account s large pools also captured larger share of rewards from the fixed economic by dumping the idena coin price and squeezing rewards from solar accounts so the number accounts number of pools increased and as they run a single note the notes as percentage of the network in turn significantly dropped uh leading to loss of throughput and security so the purpose of Delegation was to make coordinated pools visible and enable pool operators uh to handle operational hustles without knowing the private keys of the participants accounts but oddly large pools continue to show the signs of third party key access with automated simultaneous or

sequential transactions like delegations account terminations and funneling rewards earn before U uh before delegation or after termination of accounts and looking at these factors We examined top 31 pools um that had ever been delegated to more than 100 accounts in the Iden protocol history and all these top 31 pools U showed the signs of third party key access moreover some pools had um Financial ties and as we treat them as the same entity then these 31 pools were in fact only 23 entities so the peteran crisis reached it Peak in May 2022 there were only 23 entities representing less than 1% of

the distinct entities controlling uh at least 40% of the accounts and almost half of the protocol rewards this prompted us to change the model to subliner Identity staking which is the combination of proof of stake and proof of proof of personhood but this is the subject for the further research thank you thanks okay so mea's damning statistic on less than 1% of the entities controlling the keys of almost uh 40% of the Network's accounts to many in this room to most here this is uh you know sufficient condition for or sufficient

evidence P fasy evidence for puppeteering not your keys not your coins kind of encapsulates this perspective but if we were to be charitable and take the operators perspective the pool operators perspective thirdparty key control could be you know a voluntary Choice by high high information participants presuming they could hold account hold oper ators accountable offchain uh however addressing this perspective uh and looking at the evidence we observed a lack of disputes and complaints uh as well as a lack of marketing uh of pool operators uh that would evidence accountable uh custody relationships uh also notably the

jurisdictions of the large pools uh lacked rule of law Indonesia Russia and Egypt and then there were of course other factors there are these messages from the largest pool operators there was a meteor rise and fall of accounts uh So based off of this totality of facts uh you know we concluded that the uh uh excuse me uh we concluded that the model of one person one vote one reward uh had simply collapsed into many puppeteered subn networks and yet our analysis was still very cursory it excluded uh other large pools between 15 and 100 accounts it also excluded family pools with less than 15 accounts so the statistics could only get worse not better so I want to kind of take a step

back and talk about the major takeaways of this experiment and again this is a first empirical study uh of a proof of personhood protocol and proof of personhood is increasingly being touted as a solution to to all these problems around governance and economic distribution the first takeaway is that by giving humans economic incentives to periodically differentiate themselves from Bots even as low as $2 to $14 every few weeks uh the idena protocol had actually given more informed resourceful humans Financial incentives to Puppeteer less informed humans like Bots and when participants traded their time for a cha for a paycheck from more uh resourceful humans they ultimately

transformed A system that was intended to be one person one vote into one token one vote where plutocrats and puppeteers gained out ized influence now these economies of scales for the more resourceful to exploit is not something unique to idena but common to all protocols all distributed protocols um as we mentioned at the start of this talk other protocols validate humans in different ways for example biometric scans but just because humans don't necessarily have to run their own node or do periodic cognitive tests that doesn't mean that they don't have hassles that means they just have a different set of hassles which intermediaries and operators will be happy to step in and readily perform

which takes us to takeway four these exploits might not always manifest as puppeteering but can also manifest as account trading now as I mentioned at the start of the talk Ida idena styed account trading through novel mechanisms notably identity staking but other protocols such as worldcoin uh this past summer has had uh documented cases of account trading we don't know the extent to which uh so account trading in protocols should not be treated as evidence of advanced mechanisms or protections but to the contrary account trading May signal a lack of them and be a precursor to puppeteering which takes us to takeaway uh four

um the challenge of filtering humans from Bots controlling accounts can't be separated from the information challenge proof of personhood seeks to differentiate humans from Bots not their biases but as idena demonstrated when given incentives to differentiate themselves from Bots humans also have incentives to align control and Puppeteer other humans like Bots to amplify their biases and it's biases or the problem of faction that we care about when we're creating systems of excuse me Democratic governance this information challenge will only get

worse and more acute as humans integrate biologically with neural interfaces the distinction between filtering humans from Bots and humans acting like programmable Bots will blur more if not collapse revealing a more foundational challenge than establishing biological uniqueness establishing the informational uniqueness of participants or the extent to which they cluster with the same interests or biases biases now this is not a technical problem but a social one de facto sibles or puppets or humans acting like Bots are the natural objects of colluders or puppeteers by extension deao Cil resistance is a mutually implicated or mirror challenge

to collusion resistance neither can be solved independently but both must be tackled simultaneously and finally the last takeaway here is that thwarting onchain vote buying with novel mechanisms like like receipt freeness or advancments and like proofs of complete knowledge these don't actually solve for offchain vote buying into de meat space but actually may encourage it as a lowcost alternative recall idena actually undermined account training with identity stakings so the next best alternative for vote buying became offchain puppeteering or buying participants time similarly when onchain vote buying and trusted execution environments become costly we can simply

expect the more resourceful to move offchain so the Ida experiment underscores the importance of thinking in terms of incentives systems interacting systems and most importantly acknowledging the ties that arise from both talking and trading in future work we examine the relationship between incenting unique accounts and stake in a paper called entitled between zero and one and after that we explore social identity as a richer identity system in Beyond 01 thank you [Applause] do you have time for questions yes so congratulations for such a inspiring and good presentation

we thank you very much and now we have some time for some two questions does anyone have a question okay thank you I take the microphone to you hi okay um so first of all it's very impressive that you had the data to do an empirical study of how these protocols work and I wanted to ask you what was the most surprising result that uh you know I I can understand it's not easy to model and predict human behavior but maybe you had already an idea of how

it was going to work so what was the most surprising result from the empirical data on uh social coordination and then a question for M um would you do anything different if you could go back on uh the protocol knowing then what happened yeah I think the most surprising result was so so two sorry two years ago I I co-authored this paper called decentralized society and we offered a critique of proof of personhood in about two paragraphs which basically made these points um and then folks kind of didn't take it seriously and I was like all right well let's actually do the empirical work and like substantiate the claims and I was really surprised actually the extent to which there was puppeteering so we only

studied like the top 31 pools right and there was like over 400 pools but those 31 pools constituted you know about 40% of the counts right and that was puppeteered and that didn't even look at you know the rest of the pools but when you looked at kind of in the paper we show um the distribution of stake and Rewards in these other pools and they show the same pattern so we didn't go ahead and do the chain analysis but there's pretty much a strong indication that probably most of the network was Puppeteer though we can't claim that so I think the the breadth and depth is what was surprising to me yeah there was another question right yeah um if if if I if I would start it

like the protocol again maybe I would start it with a sublimal identity staking as a starting point because we don't have like much um Primitives in the in in the web three space we have only token voting and we introduce like a proof of personhood but uh anyways you you you should start and then see how it goes right like uh it's a matter of like experimentation and like empirical uh study and let's see how it goes yeah all right um yeah so first of all thanks for the presentation and I'm wondering how does the uh competitive landscape look for you and like what are some of the Alternatives because I mean maybe I can't discern the difference between

like proof of uh personhood or proof of uh humanity and other things so maybe you can summarize on that as well I mean I think so proof of humanity any any kind of like protocol that seeks to have one person one account and achieve this traditional civil resistance deur civil resistance is vulnerable to this critique I think the set of tools and set of like representations need to be expanded and particularly representing groups and relying on a web of trust uh to as a complement to Simply verifying unique

humans if you over index and solve and create incentives to solve uh one aspect like verifying biological uniqueness and it invites this information problem where people have an incentive to control humans by controlling their information so the best kind of Remedy we talk about this in this in the paper is actually to expand the identity space beyond simple biological representations to representing memberships of groups so if there's an analogy there's you know representing individuals right there if you think about social coordination there's individuals there's communities and civil society and then there's nation states and proof of personhood is

kind of trying to just solve this representing individuals you know Sovereign individuals by and then like a lot of kyc nation state stuff tries to you know offer you know centralized authentication and the thing that has been largely ignored is this middle level of civil society and groups uh which actually is the best protection against you know uh author Aran or top- down control and also you know the Anarchy of just having a bunch of um you know Sovereign individuals that uh don't have right aren't operating within social structures so actually using social structures as a way to complement uh identity I think is is the way to

move forward thank you yeah yeah agree yeah okay thank you thank you very much pan Miku thank you